# Alternatives to LangGraph > Source: https://theagentsindex.com/langgraph/alternatives (current datasheets, evidence-ranked) Candidates share the baseline category or are previously researched peers. Requirements, preserved strengths, explicit trade-offs and research coverage decide the order; paid tier does not. ## Shortlist ### 1. LangChain Open source agent frameworks plus LangSmith, a hosted platform to trace, evaluate, deploy and govern agents. **Research coverage:** 34/35 criteria **Where it improves on LangGraph:** - [Done-for-you service: partial — You run the software yourself on every self-serve plan; only Enterprise buyers get LangChain engineers, architectural guidance and team training alongside the product.](https://www.langchain.com/pricing) - [Agent permissions: yes — Human-in-the-loop interrupts gate agent actions in the framework, dcode gates code execution behind approval modes, and Fleet agents can run as themselves or as the calling user.](https://www.langchain.com/langgraph) - [Admin controls: yes — Admins cap model spend and request rates through the gateway on every plan and set monthly compute ceilings on Engine; fine-grained role and tag policies need Enterprise.](https://docs.langchain.com/langsmith/llm-gateway) **Trade-offs against LangGraph:** - [Self-hosted: partial — The open source frameworks run anywhere, but self-hosting or hybrid-hosting the LangSmith platform on Kubernetes is an Enterprise-plan option; Developer and Plus are cloud-only.](https://www.langchain.com/pricing) - [Open source: partial — The frameworks and the dcode coding agent are MIT-licensed and free, while the LangSmith platform that carries the price, seats and governance is proprietary and hosted.](https://www.langchain.com/langchain) - [Bring your own key: partial — You can bring your own provider keys through the gateway on every plan, or buy inference as Gateway Credits, but Engine refuses your keys and bills its model use as compute units.](https://docs.langchain.com/langsmith/engine) [View LangChain](https://theagentsindex.com/langchain) ### 2. Mastra Open-source TypeScript framework for building AI agents, with a hosted platform for running and observing them. **Research coverage:** 35/35 criteria **Where it improves on LangGraph:** - [Done-for-you service: partial — You run the software, but Mastra sells professional services (Concierge, forward-deployed engineering) and enterprise support covering planning, prototyping and workshops.](https://mastra.ai/terms-of-service) - [Agent permissions: yes — Tool approval can gate any tool call, per server or by predicate on tool name/arguments; approvals propagate up a subagent delegation chain.](https://mastra.ai/ai-agent-framework) **Trade-offs against LangGraph:** - [In your pipeline: partial — Scorers and eval suites run in CI through Vitest/Jest via runEvals; the vendor claims no agent that acts on a pull request.](https://mastra.ai/docs/evals/running-in-ci) - [Seat model: no — Not licensed per seat: Starter and Teams both state unlimited users, and the self-hosted Enterprise offer explicitly rejects per-seat metering.](https://mastra.ai/pricing) - [Certifications: partial — SOC 2 documentation is offered from the Teams plan; a DPA covering GDPR, UK GDPR and CCPA/CPRA is published, and a Vanta-hosted trust centre carries the security programme. No ISO 27001 claim found.](https://mastra.ai/pricing) [View Mastra](https://theagentsindex.com/mastra) ### 3. Agno Python SDK, open-source AgentOS runtime and hosted control plane for building and operating agents. **Research coverage:** 35/35 criteria **Where it improves on LangGraph:** - [Debugs and fixes: yes — Bug fixing is a documented pattern: reproduce, patch, rerun the failing tests.](https://docs.agno.com/use-cases/coding-agents) - [In your editor: yes — `agno connect` writes MCP config for Claude Code, Claude Desktop, Codex and Cursor so those clients can run your agents.](https://docs.agno.com/cli/connect) - [Admin controls: yes — JWT scopes gate every endpoint per agent and per tool, with roles assigned in the control plane and guardrails at runtime.](https://docs.agno.com/features/security-and-auth) **Trade-offs against LangGraph:** - [In your pipeline: no — The enumerated serving surfaces are REST, MCP in Claude/ChatGPT and chat apps; no pull-request or CI surface appears.](https://www.agno.com/) - [Open source: partial — SDK and AgentOS are Apache 2.0; the control plane is the commercial part and is not published as source.](https://www.agno.com/compare/agno-vs-langgraph) - [Pooled budget: partial — There is no metered balance to pool; seats share access to the same connected runtimes and usage inside a connection is unlimited.](https://www.agno.com/pricing) [View Agno](https://theagentsindex.com/agno) ### 4. CrewAI Open-source multi-agent framework with a managed platform for deploying and governing agent workflows. **Research coverage:** 34/35 criteria **Where it improves on LangGraph:** - [Done-for-you service: partial — You run the software yourself, but the Enterprise plan sells forward-deployed engineers, onboarding and training separately, so people-delivered build help exists as a paid extra.](https://www.crewai.com/pricing) **Trade-offs against LangGraph:** - [In your pipeline: partial — Deployments connect to GitHub repositories and agents can file issues and cut releases, but the documented action list contains no pull-request review or CI job step.](https://docs-platform.crewai.com/platform/en/integrations/github) - [How it meters: partial — Metering is per workflow execution rather than per seat, and the free tier states its allowance, but Enterprise volumes are sized case by case with unpriced overage.](https://www.crewai.com/pricing) - [Pooled budget: partial — The published allowance is stated per plan per month rather than per member, so usage is drawn from one plan-level pool, but no per-seat or per-team budget split is documented.](https://www.crewai.com/pricing) [View CrewAI](https://theagentsindex.com/crewai) ### 5. Letta Stateful agent harness, cloud and SDK for agents with persistent, git-versioned memory. **Research coverage:** 35/35 criteria **Where it improves on LangGraph:** - [Debugs and fixes: yes — Vendor states coding agents debug failures and run test suites.](https://docs.letta.com/handbook/index.md) - [Agent permissions: yes — Four modes from strict to unrestricted, plus allow/deny tool patterns and a cross-agent memory guard.](https://docs.letta.com/configuration/permissions/index.md) - [In your editor: yes — The letta-acp adapter runs Letta as a custom agent in Zed, JetBrains IDEs and Obsidian.](https://docs.letta.com/platform/acp/index.md) **Trade-offs against LangGraph:** - [MCP server: no — MCP runs inbound only: Letta sessions consume stdio, HTTP and SSE MCP servers; no Letta MCP server is published.](https://docs.letta.com/agent-sdk/mcp/index.md) - [Seat model: partial — Priced per seat on Teams Pro; no seat minimum or maximum is published, and Enterprise seat pricing is by contact.](https://docs.letta.com/pricing/index.md) - [Pooled budget: no — Quota attaches to each purchased Pro seat; unpaid seats get no Letta-hosted inference at all.](https://docs.letta.com/teams/quickstart/index.md) [View Letta](https://theagentsindex.com/letta) ### 6. AgentsKit Composable MIT-licensed TypeScript packages for agent runtime, tools, memory, RAG and chat UI. **Research coverage:** 34/35 criteria **Where it improves on LangGraph:** - [Agent permissions: yes — Any tool can be marked as requiring confirmation, which pauses the run and waits for an explicit approve or deny, and a sandbox policy layer adds allow/deny lists on top.](https://www.agentskit.io/docs/agents/hitl) - [Admin controls: yes — A mandatory sandbox wraps every tool with allow lists, deny lists and per-tool validators, and the policy is snapshotted so later callers cannot widen it.](https://www.agentskit.io/docs/production/security/mandatory-sandbox) - [Audit log: yes — Runs can be recorded to a hash-chained, HMAC-signed log whose verify() throws on tamper; it is a library you attach as an observer, not a hosted admin console.](https://www.agentskit.io/docs/production/observability/audit-log) **Trade-offs against LangGraph:** - [In a browser: partial — Headless chat bindings for seven frameworks let you build a web surface, but the hosted product chat experience is a separate sibling site rather than part of this library.](https://www.agentskit.io/llms.txt) - [How it meters: partial — AgentsKit itself bills nothing: your only spend is the model provider's tokens, which the cost-guard observer caps with a hard dollar ceiling per run.](https://www.agentskit.io/llms.txt) - [A company can buy it: partial — There is nothing to buy for the library itself; the managed product AKOS is the commercial path and is sales-led.](https://www.agentskit.io/docs/production) [View AgentsKit](https://theagentsindex.com/agentskit) ## Method Unknown data is never treated as a benefit or a trade-off. Featured placement is labelled but cannot alter recommendation order.